Abstract:Global navigation satellite system(GNSS) coordinate series accuracy is mainly affected common mode error(CME) influence. In order to improve the accuracy of GNSS coordinate series, this paper adopts variational bayesian independent component analysis(vbICA) method to extract CME of coordinate series of 20 GNSS stations in the experimental site, and uses distance correlation coefficient and root mean square(RMS) as indicators to evaluate the filtering effect of the original coordinate series. The filtering performance of vbICA method is compared with PCA and ICA. The results show that the filtering effect of vbICA is obviously better than PCA and ICA. After vbICA filtering, the RMS of residual coordinate series in E,N,U direction decreases by 36.57%, 31.63% and 10.97% on average. Distance correlation coefficient decreased by 60.53%,56.84% and 25.80% on average. Considering the optimal noise model and excluding CME, GNSS velocity field estimation is more reliable and accurate, effectively improving GNSS coordinate series accuracy, and providing reliable data support for geodynamic research.